feat: extract_layout implementation for native Poseidon2

[stephenh-axiom-xyz]

Resolves INT-4288.

Discussed with @arayikhalatyan, for trace generation purposes we can actually treat native Poseidon2 as if it uses fixed-size records. This simplifies the GPU tracegen and allows us to not send any auxiliary information to the GPU device.

[codspeed-hq]

CodSpeed WallTime Performance Report

Merging #1813 will not alter performance

_{Comparing feat/native-poseidon2-extract (5c9157f) with feat/new-execution (040d653)}

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

Summary

✅ 20 untouched benchmarks

[github-actions]

group	app.proof_time_ms	app.cycles	app.cells_used	leaf.proof_time_ms	leaf.cycles	leaf.cells_used
verify_fibair	(+25 [+2.0%]) 1,260	322,648	17,270,578	-	-	-
fibonacci	(+6 [+0.3%]) 2,358	1,500,277	50,589,231	-	-	-
regex	(+56 [+0.8%]) 6,982	4,165,432	166,449,586	-	-	-
ecrecover	(+24 [+1.7%]) 1,416	137,260	8,174,585	-	-	-
pairing	(-47 [-1.2%]) 3,984	1,862,964	102,530,853	-	-	-

Commit: 5c9157f

Benchmark Workflow

[arayikhalatyan]

I believe this idea should work, and even makes it much easer to do parallel tracegen over rows on GPU.
Note, this is different from other chips in the sense that we will 'decode' the records from buffer differently from how they were 'allocated'. During execution we allocate 1 record (aka NativePoseidon2RecordMut) per instruction but in buffer decoding during tracegen we will treat it as 1 record (aka &mut NativePoseidon2Cols) per row